改进的加权极速学习机  

Improved Weighted Extreme Learning Machine

作　　者：邢胜[1,2] 王晓兰[3] 赵士欣[1,4] 赵彦霞[1,5] 

机构地区：[1]河北大学管理学院,保定071002 [2]沧州师范学院计算机科学与工程学院,沧州061001 [3]沧州职业技术学院信息工程系,沧州061001 [4]石家庄铁道大学数理系,石家庄050043 [5]河北经贸大学信息技术学院,石家庄050061

出　　处：《计算机科学》2017年第4期275-280,共6页Computer Science

基　　金：国家自然科学基金资助项目(71371063;61170040;61672205);河北省人力资源社会保障科研合作课题(JRSHZ-2016-07038)资助

摘　　要：针对加权极速学习机人为固定权重可能会错失更优权重的问题,提出了改进的加权极速学习机。该方法的多数类的初始权重设为1,使用多数类与少数类样例数的比值作为少数类的初始权重,然后通过在多数类或者少数类中添加权重调节因子,从缩小和扩大两个方向去调节权重,最后通过实验结果选出最优的权重。实验分别使用原加权极速学习机、其他权重的极速学习机和新方法在改造的UCI数据集上进行比较。结果表明新方法无论是在F-measure还是G-mean上都要优于其他加权极速学习机。If the weight of the original weighted extreme learning machine is fixed artificially,the more optimal weight may be missed.Aiming at this problem,an improved weighted extreme learning machine was proposed.The new method uses the ratio of the sample number of different classes as the initial weight.The weight is adjusted by the weight adjustment factor from two directions of reducing and enlarging the weight ratio.Finally,the optimal weights are selected by the experimental results.The experiments were carried out on the transformed UCI data set using the original weighted extreme learning machine,the weighted extreme learning machine with other weights and the new method respectively.The experimental results indicate that the improved weighted extreme learning machine has better classification performance.

关 键 词：不平衡学习 加权极速学习机 代价敏感学习 单隐层前馈网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]